EVOTS: Evolutionary Transformer Search for Time Series Forecasting
EVOTS: Evolutionary Transformer Search for Time Series Forecasting
EVOTS：用于时间序列预测的进化 Transformer 搜索
Evolutionary neural architecture design for multivariate time-series forecasting remains underexplored, with most approaches relying on fixed Transformer architectures despite substantial variation across tasks and forecasting settings. 针对多变量时间序列预测的进化神经网络架构设计目前仍未得到充分探索，尽管不同任务和预测设置之间存在显著差异，但大多数方法仍依赖于固定的 Transformer 架构。
This paper introduces an evolutionary neural architecture search framework for discovering task-adaptive Transformer-like models for time-series forecasting (EVOTS). 本文介绍了一种进化神经网络架构搜索框架，旨在为时间序列预测发现任务自适应的类 Transformer 模型（EVOTS）。
Architectures are encoded using a modular genome representation that enables flexible composition of attention, feed-forward, and projection components, while a repair mechanism enforces structural validity throughout the evolutionary process. 该架构使用模块化基因组表示进行编码，能够灵活组合注意力机制、前馈网络和投影组件，同时通过修复机制在整个进化过程中强制执行结构有效性。
This formulation allows effective exploration of a diverse architecture space without relying on hand-crafted design rules. 这种表述方式使得在不依赖手工设计规则的情况下，能够有效地探索多样化的架构空间。
The proposed approach is evaluated on four benchmark datasets from the ETT family (ETTh1, ETTh2, ETTm1, and ETTm2) under multiple forecasting settings, including univariate-to-univariate, multivariate-to-univariate, and multivariate-to-multivariate prediction, with horizons of 96, 192, 336, and 720. 该方法在 ETT 系列的四个基准数据集（ETTh1、ETTh2、ETTm1 和 ETTm2）上进行了评估，涵盖了多种预测设置，包括单变量到单变量、多变量到单变量以及多变量到多变量预测，预测步长分别为 96、192、336 和 720。
In the multivariate-to-multivariate setting, the evolved architectures achieve competitive and, in several cases, improved mean squared error relative to a strong Transformer-based baseline. 在多变量到多变量的设置中，进化出的架构与强大的 Transformer 基准模型相比，实现了具有竞争力的均方误差，在某些情况下甚至表现更优。
Additional analyses examine performance differences across forecasting settings and report wall-clock training time to provide a coarse indication of computational cost. 额外的分析考察了不同预测设置下的性能差异，并报告了实际训练时间，以提供对计算成本的粗略评估。
Overall, the results demonstrate that evolutionary search can effectively discover flexible and high-performing Transformer-like architectures for multivariate time-series forecasting within practical runtime constraints. 总的来说，研究结果表明，进化搜索能够在实际的运行时间限制内，有效地为多变量时间序列预测发现灵活且高性能的类 Transformer 架构。